---
title: 'In the News: October 8, 2026 (Extra 2)'
description: "Anthropic's Haiku 5.5 cuts small-model prices about 75% and halves Sonnet 5.5 cache-read prices, which lowers the cost of subagent work in coding harnesses."
canonical_url: 'https://darkfactory.dev/news/2026-10-08-extra-2'
markdown_url: 'https://darkfactory.dev/news/2026-10-08-extra-2.md'
collection: news
date_published: '2026-10-08T16:10:00-04:00'
date_modified: '2026-10-08T16:10:00-04:00'
---

# In the News: October 8, 2026 (Extra 2)


One item this edition: Anthropic released Claude Haiku 5.5 and cut Sonnet 5.5 cache-read prices on the same day. Both change what a subagent, a compaction step or a long agent loop costs to run.

## 1. Anthropic ships Haiku 5.5 at a fraction of Haiku 4.5's price and halves Sonnet 5.5 cache reads

**[Introducing Claude Haiku 5.5](https://www.anthropic.com/claude-haiku-5-5)** · Anthropic · October 7, 2026

Anthropic describes Haiku 5.5 as its "cheapest, fastest, and most capable small model" and says it pairs with Opus 5.5 and Sonnet 5.5 "as a subagent on coding work." Per million tokens, for prompts up to 100,000 tokens, Anthropic lists input at $0.10 and output at $0.50, against $1.00 and $5.00 for Haiku 4.5. Anthropic puts the average saving at around 75%, and notes the new tokenizer uses slightly more tokens per task. The model has an adjustable effort setting, a first for a Haiku-class model.

Separately, Anthropic cut Sonnet 5.5 cache reads from $0.20 to $0.10 per million tokens. It says that lowers the cost of Sonnet 5.5 on most agentic tasks by around 20%, because cache reads make up a large share of token use.

Anthropic's own benchmark table shows where the limit is. On Terminal-Bench 4.0, Haiku 5.5 scores 39.2% against 70.6% for Sonnet 5.5. On FrontierCode 1.1 (Main) it scores 46.4% against 52.1%. Anthropic says Sonnet 5.5 and Opus 5.5 "remain better choices for complex agentic coding tasks," and that Haiku 5.5 suits narrowly scoped work such as compaction, summarization and subagents. Cognition's Walden Yan says Haiku 5.5 as the sidekick in Devin Fusion holds a FrontierCode score of 66.2 while cutting cost and latency. All figures are Anthropic's or its customers' own; the customer quotes were chosen for the announcement.

**Why it matters:** If your harness sends every step to one large model, the price list now gives you a reason to split the work. Test Haiku 5.5 on compaction, lookups and bounded subagent tasks, and keep the larger model on the steps that decide whether the code works. The Sonnet 5.5 cache-read cut helps long loops with no change on your side. The Terminal-Bench gap says a verification loop should still sit between a small model's output and your main branch.